I don’t know if this has come up before or not, but you know what would be a cool application of AI for D&D? An art generator trained on WotC’s art for the playable races that you can use to generate character portraits. If they keep the art in-house they can hardly be accused of stealing, and it would make it easier to get something like a Goliath or Genasi or Thri-Keen than one trained on general terms. No idea if it’s practical, but it’d be something they could market to both sides of the table.
Nowhere near enough art to do it with AI. They could do it without AI, in exactly the way video games do it, and I suspect they're working on it for the VTT.
Building on the above, the reason AI dredges other content is the scales involved. Wizards could upload every single piece of art from every single D&D book. Every single piece of art from every single Magic printing (28,338 distinct cards, many with multiple different arts). Every single piece of art from all the other games published under their umbrella.
Wizards can pull together tens of thousands of pieces of distinct art, but it would be a drop in a bucket compared to the millions of pieces of data generative AI require to function. And that is before considering how much art styles within Wizards’ data points differ - each edition has its own artistic feel; Magic has all kinds of different art styles; some of the other Wizards-published games are not even in the same genera, etc.
Now, Wizards could add their art to an existing AI’s data pool and probably train that AI to recognize certain Wizard words like “Tiefling” and such, so it can create a facsimile of a Tiefling built in part on Wizards’ art. But, if a generative AI system like that is to work, it will take a whole lot more than what Wizards alone can provide.
I don’t know if this has come up before or not, but you know what would be a cool application of AI for D&D? An art generator trained on WotC’s art for the playable races that you can use to generate character portraits. If they keep the art in-house they can hardly be accused of stealing, and it would make it easier to get something like a Goliath or Genasi or Thri-Keen than one trained on general terms. No idea if it’s practical, but it’d be something they could market to both sides of the table.
Nowhere near enough art to do it with AI. They could do it without AI, in exactly the way video games do it, and I suspect they're working on it for the VTT.
they could start a bounty program. start a monthly contest that winnows down something like 40 entries a month to some smaller number of entries via votes from a poll. computer filters out very samey images (or pallet-swaps, same pic different medium, duplicates, nudity, politics, etc) to make it a varied contest. monthly bonus bounty on whatever race/species/class is least represented, separate bracket for ten or so of those. pit four or five at a time against each other in a ranked choice vote battle with comments turned on. artist's name shows up when you hover their latest submission, but also shows other submissions. commenters would potentially (hopefully) catch out stolen art and repeat offenders during the week, steer votes away from cheaters. winners go into a DBB expanded character portraits database (no or limited external linking to reduce database load?). winners (and some hand-picked submissions) would go into the AI training algorithm soup. once it's up and running, the AI could customize forums and character portraits like a fingerprint/stamp/signature in lieu of signature on future d&d type submissions (fan art, book art, DM's Guild products, etc).
...as for the bounty? can't be real money or it'll go the way of temu or coin mining. most winning winners obviously show up on some ranking page. there they'd have the option to link their personal art page or patreon or whatnot. also, maybe a one-use DBB discount code (because artists need to buy books too).
and every artist in the AI soup is there with their consent (which they could petition to revoke at any time). that would be the up-front mission from the beginning: training AI with consent for low-res digital only thumbnail art generation which (hopefully) wouldn't threaten to replace actual fleshy meat-space artist people.
even with current tech you can get really good LoRAs from five photos, that allow a good amount of change. getting tokens to train a LLM is very different than using LoRAs or IP-adapters to get a cool image that you can turn into a mini or digital asset.
i would love to have an AI assistant to help with encounters, backstory... hell, let me ask AI about some lore that i don't want to BS or spend 10 minutes reading into live at the table.
Honestly if WotC wants to do what current "ai" is offering/capable of to have an "ai" DM there are better options than "ai" can deliver with what WotC owned IP's current catalog has to offer, which is tiny compared to what it takes to properly "train" the current "ai" options not to mention the unknown responses "ai" will have to questions that will ultimately turn into PR nightmares for WotC. The money it would cost to make a sub par "ai" DM could be used far more effectively with other tech currently available to make either better tools for DM's or more content for all players. That said there is nothing wrong with keeping an eye on future tech, but what will they have to sacrifice to be "cutting edge" in the current "ai" tech, likely less than is profitable and useable for the current paying players.
AI Art, the ability to generate sounds, getting the AI to translate old written content to a more modern and readable format… These are all AI tools that I use daily in my projects.
But the people who want to use AI in D&D want it to be used as a way to take away the creative part of a DM’s job
I have to ask... why is it OK for you to use AI in your projects that "takes away the creative part", but not WotC?
Rollback Post to RevisionRollBack
Active characters:
Edoumiaond Willegume "Eddie" Podslee, Vegetanian scholar (College of Spirits bard) Lan Kidogo, mapach archaeologist and treasure hunter (Knowledge cleric) Mardan Ferres, elven private investigator (Assassin rogue) Peter "the Pied Piper" Hausler, human con artist/remover of vermin (Circle of the Shepherd druid) Xhekhetiel, halfling survivor of a Betrayer Gods cult (Runechild sorcerer/fighter)
if the goal is "zork but more options," that's not an AI issue. it wouldn't be easy to turn something like waterdeep dragon heist into a choose-your-own-adventure that takes into account several players and their combination of character classes and backgrounds, but it's also not beyond the pale to imagine paid professionals doing that if tasked with it. all it takes is a bigger cassette tape and some code customization... which very broadly describes BG3. they could have sold BG3 as a live party members experience with maps and webcams and voice chat and it still wouldn't have been AI.
Rollback Post to RevisionRollBack
unhappy at the way in which we lost individual purchases for one-off subclasses, magic items, and monsters?
tell them you don't like features disappeared quietly in the night: providefeedback!
if the goal is "zork but more options," that's not an AI issue.
It's absolutely an AI issue. It might not be a generative AI issue, unless you're trying to incorporate chat (side note: I've seen someone implement Eliza in a text-based mud back in the 1990s and fool a surprising number of people), but any time you've got entities on screen that are supposed to represent creatures, they've got a controlling AI (though potentially very primitive).
if the goal is "zork but more options," that's not an AI issue.
It's absolutely an AI issue. It might not be a generative AI issue, unless you're trying to incorporate chat (side note: I've seen someone implement Eliza in a text-based mud back in the 1990s and fool a surprising number of people), but any time you've got entities on screen that are supposed to represent creatures, they've got a controlling AI (though potentially very primitive).
a clever algorithm isn't AI, although an AI is a stack of clever algorithms. personally, i wouldn't mind trying out an official d&d module via algorithm (especially if the dm persona on screen was a collaboration 'deep fake' likeness of patrick rothfuss, ed greenwood, gandalf the grey, etc!!). an efficient program on a big database of text strings that simulates innovation and varied responses (like baldur's gate 3) is just the sort of thing a big corporation would approve since it wouldn't be convinced (read: trained by user interaction) to condone racism, sell you a car for a dollar, reveal passwords, launch nukes, etc if it lacked machine learning.
(edit: the original post and the CEO of Hasbro are using 'AI' to refer to more advanced generative AI. the Bigby's artist was controversial because they used an AI to generate new content, not because their photoshop's blur tool was too complex. even post-AI society in Dune still used machines and algorithms, just not "thinking machines." )
Video game algorithms are absolutely referred to as AI. It's a specialized field that doesn't have a whole lot to do with things like modern generative AI, but it's still called AI.
Video game algorithms are absolutely referred to as AI. It's a specialized field that doesn't have a whole lot to do with things like modern generative AI, but it's still called AI.
from your link: "whereas 'real AI' addresses fields of machine learning, decision making based on arbitrary data input, and even the ultimate goal of strong AI that can reason, 'game AI' often consists of a half-dozen rules of thumb, or heuristics, that are just enough to give a good gameplay experience."
artists and creators today aren't concerned by 'game AI' on the rise. legitimate concerns about AI shouldn't be diluted by whether or not zork is okay. thus, zork zork zork. :6
Rollback Post to RevisionRollBack
unhappy at the way in which we lost individual purchases for one-off subclasses, magic items, and monsters?
tell them you don't like features disappeared quietly in the night: providefeedback!
Seriously no one is actually using a real AI today and there is no real Ai any place close to being ready.
What everyone on line is doing is fooling with fancy search engines. Or they are just playing around to try to break it.
What D&D could use is a good D&D only search engine. If a dm or player has a question put it to the search engine and official answers will come out. It could give out the RAW and anything else they want to. It could even be limited to purchased content.
They could even make a campaign book specific system. Each source book gets audio descriptions of locations, buildings and even rooms. Anything to help the DM. There could even be world or location specific equipment lists with different costs and availability for all the equipment and magic items. The list could be adjustable by the Dm.Then no one needs to wonder how much something costs in Water Deep vs any other area in any other world. Maps could have travel times already calculated out by the route you want to take.
All without some type of AI running a single mission. All something the DM would pay for.
Smartphone screens are not the toy industry’s only technology challenge. Cocks uses artificial intelligence tools to generate storylines, art, and voices for his D&D characters and hails AI as “a great leveler for user-generated content.”
Current AI platforms are failing to reward creators for their work, “but I think that’s solvable,” he says, describing himself as “an AI bull” who believes the technology will extend the reach of Hasbro’s brands. That could include subscription services letting other Dungeon Masters enrich their D&D campaigns, or offerings to let parents customize Peppa Pig animations. “It’s supercharging fandom,” he says, “and I think that’s just net good for the brand.”
Cocks’ company attracted the attention of one AI platform owner last November, when a social media debate about whether D&D had gone “woke” prompted Elon Musk to ask, “How much is Hasbro?”
Cocks laughs off the tweet from the world’s richest man, saying any takeover interest went nowhere.
Rollback Post to RevisionRollBack
Met D&D in 1976 while in the Navy, learning Russian, and fell in love. We were steady for several years, then drifted apart and saw other people. Ran into each other a few years ago and the spark kinda rekindled, ya know? It’s an open relationship and I hang with GURPS and DragonQuest, but we’re all mutuals and it’s good. Open communication is the real key to any polydice, it’s the only real rule.
There might not be a formal definition, but when someone says "real AI" they typically are referring to a hypothetical actual intelligence like R2-D2 or Data, who are fully sapient beings capable of abstract reasoning and expressing their own desires and ideas. That's often what techbros sell AI as heading towards, too, even though it absolutely isn't.
Rollback Post to RevisionRollBack
Find your own truth, choose your enemies carefully, and never deal with a dragon.
"Canon" is what's factual to D&D lore. "Cannon" is what you're going to be shot with if you keep getting the word wrong.
Real AI is both artificial and, more to the point, intelligent. It doesn't have to be human level sapient or anything, just capable of understanding what it's communicating. A dog understands that when it barks, it is communicating to us that there is a potential intruder. A cat understands that if a mouse runs into a tube, it has to come out either end and will start planning accordingly.
What's called AI at the moment is not intelligent. It can trick you into thinking that it understands, but really it's just throwing words out that make sense. Sometimes. Thus it's not true AI.
Rollback Post to RevisionRollBack
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Intelligence is fundamentally about what's going on under the hood. If you can't accept discussions involving that, then the discussion is at a dead end.
Rollback Post to RevisionRollBack
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Ed Zitron has a great piece about how it's all a con to refer to things like ChatGPT as "AI."
And others about its impact on labour as well as on the environment.
The dissonance going on among those who claim to take protecting the latter seriously only to be cheering and applauding the development and use of generative AI is a dissonance becoming those in cults.
Intelligence is fundamentally about what's going on under the hood. If you can't accept discussions involving that, then the discussion is at a dead end.
IF that were true (and it isn’t), then I would have no basis to assume you are intelligent without first dissecting your brain and then, somehow, getting it to run after I rip your skull open.
If I don’t do that, then I don’t know what is going on under your hood, I can only make guesses.
Nonsense. I didn't any such thing. You claimed the presence of intelligence is based on performance, it's not. That doesn't mean you have to dissect someone to know if they have intelligence.
I'm not sure whether you genuinely don't understand what's being said or if this has devolved into yet another "I must prove myself right at all costs!" thing, but the former will take a lot more effort that it's really worth to correct because it means going into a deep dive on what intelligence really is and the latter...is a boring conversation.
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Intelligence is fundamentally about what's going on under the hood. If you can't accept discussions involving that, then the discussion is at a dead end.
Talking about what's going on 'under the hood' is fundamentally a dead end, because while we have some idea of what's going on on a fine detail level, at the holistic level required to evaluate whether something 'knows' what it's doing, we don't really know what's going on under the hood... for either humans or modern neural-net AIs. The reason we use observed behavior is because that's the information we have. I don't actually know that you're a thinking human being, I just assume that you are because you behave like one.
Intelligence is fundamentally about what's going on under the hood. If you can't accept discussions involving that, then the discussion is at a dead end.
Talking about what's going on 'under the hood' is fundamentally a dead end, because while we have some idea of what's going on on a fine detail level, at the holistic level required to evaluate whether something 'knows' what it's doing, we don't really know what's going on under the hood... for either humans or modern neural-net AIs. The reason we use observed behavior is because that's the information we have. I don't actually know that you're a thinking human being, I just assume that you are because you behave like one.
... and you'd only need to meet him / her in person and observe him / her to clarify what you suspect and—and this is important—what he / she knows.
Many are the works of philosophy and theology that explore what it means to be a 'thinking human being.' What you are saying about how this can only be 'observed' is not at all 'deep.' Because we are all conscious of this ourselves. What you are saying is analogous with saying some occurrence might not have occurred because no one was there to see it happen. Which is preposterous. It occurred or it didn't. An occurrence is not dependent on some observer's subjective experience of it. And someone's being a 'thinking human being' isn't dependent on it either. The universe doesn't revolve around us or around our awareness of it. We are aware of our belonging to the species to which we belong and of the fact it is a species with a capacity for intelligence. Science tells us that much. No amount of programming of mere language learning models is going to make them belong to our species. Or make them possess true intelligence. No matter how much we further ruin the environment in the process.
Building on the above, the reason AI dredges other content is the scales involved. Wizards could upload every single piece of art from every single D&D book. Every single piece of art from every single Magic printing (28,338 distinct cards, many with multiple different arts). Every single piece of art from all the other games published under their umbrella.
Wizards can pull together tens of thousands of pieces of distinct art, but it would be a drop in a bucket compared to the millions of pieces of data generative AI require to function. And that is before considering how much art styles within Wizards’ data points differ - each edition has its own artistic feel; Magic has all kinds of different art styles; some of the other Wizards-published games are not even in the same genera, etc.
Now, Wizards could add their art to an existing AI’s data pool and probably train that AI to recognize certain Wizard words like “Tiefling” and such, so it can create a facsimile of a Tiefling built in part on Wizards’ art. But, if a generative AI system like that is to work, it will take a whole lot more than what Wizards alone can provide.
they could start a bounty program. start a monthly contest that winnows down something like 40 entries a month to some smaller number of entries via votes from a poll. computer filters out very samey images (or pallet-swaps, same pic different medium, duplicates, nudity, politics, etc) to make it a varied contest. monthly bonus bounty on whatever race/species/class is least represented, separate bracket for ten or so of those. pit four or five at a time against each other in a ranked choice vote battle with comments turned on. artist's name shows up when you hover their latest submission, but also shows other submissions. commenters would potentially (hopefully) catch out stolen art and repeat offenders during the week, steer votes away from cheaters. winners go into a DBB expanded character portraits database (no or limited external linking to reduce database load?). winners (and some hand-picked submissions) would go into the AI training algorithm soup. once it's up and running, the AI could customize forums and character portraits like a fingerprint/stamp/signature in lieu of signature on future d&d type submissions (fan art, book art, DM's Guild products, etc).
...as for the bounty? can't be real money or it'll go the way of temu or coin mining. most winning winners obviously show up on some ranking page. there they'd have the option to link their personal art page or patreon or whatnot. also, maybe a one-use DBB discount code (because artists need to buy books too).
and every artist in the AI soup is there with their consent (which they could petition to revoke at any time). that would be the up-front mission from the beginning: training AI with consent for low-res digital only thumbnail art generation which (hopefully) wouldn't threaten to replace actual fleshy meat-space artist people.
unhappy at the way in which we lost individual purchases for one-off subclasses, magic items, and monsters?
tell them you don't like features disappeared quietly in the night: provide feedback!
even with current tech you can get really good LoRAs from five photos, that allow a good amount of change. getting tokens to train a LLM is very different than using LoRAs or IP-adapters to get a cool image that you can turn into a mini or digital asset.
i would love to have an AI assistant to help with encounters, backstory... hell, let me ask AI about some lore that i don't want to BS or spend 10 minutes reading into live at the table.
i loved zork, i would approve an upgrade.
Zork does make for great backstory.
Honestly if WotC wants to do what current "ai" is offering/capable of to have an "ai" DM there are better options than "ai" can deliver with what WotC owned IP's current catalog has to offer, which is tiny compared to what it takes to properly "train" the current "ai" options not to mention the unknown responses "ai" will have to questions that will ultimately turn into PR nightmares for WotC. The money it would cost to make a sub par "ai" DM could be used far more effectively with other tech currently available to make either better tools for DM's or more content for all players. That said there is nothing wrong with keeping an eye on future tech, but what will they have to sacrifice to be "cutting edge" in the current "ai" tech, likely less than is profitable and useable for the current paying players.
CENSORSHIP IS THE TOOL OF COWARDS and WANNA BE TYRANTS.
I have to ask... why is it OK for you to use AI in your projects that "takes away the creative part", but not WotC?
Active characters:
Edoumiaond Willegume "Eddie" Podslee, Vegetanian scholar (College of Spirits bard)
Lan Kidogo, mapach archaeologist and treasure hunter (Knowledge cleric)
Mardan Ferres, elven private investigator (Assassin rogue)
Peter "the Pied Piper" Hausler, human con artist/remover of vermin (Circle of the Shepherd druid)
Xhekhetiel, halfling survivor of a Betrayer Gods cult (Runechild sorcerer/fighter)
zork zork zork :p
if the goal is "zork but more options," that's not an AI issue. it wouldn't be easy to turn something like waterdeep dragon heist into a choose-your-own-adventure that takes into account several players and their combination of character classes and backgrounds, but it's also not beyond the pale to imagine paid professionals doing that if tasked with it. all it takes is a bigger cassette tape and some code customization... which very broadly describes BG3. they could have sold BG3 as a live party members experience with maps and webcams and voice chat and it still wouldn't have been AI.
unhappy at the way in which we lost individual purchases for one-off subclasses, magic items, and monsters?
tell them you don't like features disappeared quietly in the night: provide feedback!
It's absolutely an AI issue. It might not be a generative AI issue, unless you're trying to incorporate chat (side note: I've seen someone implement Eliza in a text-based mud back in the 1990s and fool a surprising number of people), but any time you've got entities on screen that are supposed to represent creatures, they've got a controlling AI (though potentially very primitive).
a clever algorithm isn't AI, although an AI is a stack of clever algorithms. personally, i wouldn't mind trying out an official d&d module via algorithm (especially if the dm persona on screen was a collaboration 'deep fake' likeness of patrick rothfuss, ed greenwood, gandalf the grey, etc!!). an efficient program on a big database of text strings that simulates innovation and varied responses (like baldur's gate 3) is just the sort of thing a big corporation would approve since it wouldn't be convinced (read: trained by user interaction) to condone racism, sell you a car for a dollar, reveal passwords, launch nukes, etc if it lacked machine learning.
(edit: the original post and the CEO of Hasbro are using 'AI' to refer to more advanced generative AI. the Bigby's artist was controversial because they used an AI to generate new content, not because their photoshop's blur tool was too complex. even post-AI society in Dune still used machines and algorithms, just not "thinking machines." )
unhappy at the way in which we lost individual purchases for one-off subclasses, magic items, and monsters?
tell them you don't like features disappeared quietly in the night: provide feedback!
Video game algorithms are absolutely referred to as AI. It's a specialized field that doesn't have a whole lot to do with things like modern generative AI, but it's still called AI.
from your link: "whereas 'real AI' addresses fields of machine learning, decision making based on arbitrary data input, and even the ultimate goal of strong AI that can reason, 'game AI' often consists of a half-dozen rules of thumb, or heuristics, that are just enough to give a good gameplay experience."
artists and creators today aren't concerned by 'game AI' on the rise. legitimate concerns about AI shouldn't be diluted by whether or not zork is okay. thus, zork zork zork. :6
unhappy at the way in which we lost individual purchases for one-off subclasses, magic items, and monsters?
tell them you don't like features disappeared quietly in the night: provide feedback!
Seriously no one is actually using a real AI today and there is no real Ai any place close to being ready.
What everyone on line is doing is fooling with fancy search engines. Or they are just playing around to try to break it.
What D&D could use is a good D&D only search engine. If a dm or player has a question put it to the search engine and official answers will come out.
It could give out the RAW and anything else they want to. It could even be limited to purchased content.
They could even make a campaign book specific system. Each source book gets audio descriptions of locations, buildings and even rooms. Anything to help the DM. There could even be world or location specific equipment lists with different costs and availability for all the equipment and magic items. The list could be adjustable by the Dm.Then no one needs to wonder how much something costs in Water Deep vs any other area in any other world. Maps could have travel times already calculated out by the route you want to take.
All without some type of AI running a single mission. All something the DM would pay for.
Chis Cock is still talking about AI:
https://www.semafor.com/article/03/06/2025/we-had-to-go-back-to-play-how-hasbros-dungeon-master-ceo-got-serious-about-games
‘AI is a great leveler’
Smartphone screens are not the toy industry’s only technology challenge. Cocks uses artificial intelligence tools to generate storylines, art, and voices for his D&D characters and hails AI as “a great leveler for user-generated content.”
Current AI platforms are failing to reward creators for their work, “but I think that’s solvable,” he says, describing himself as “an AI bull” who believes the technology will extend the reach of Hasbro’s brands. That could include subscription services letting other Dungeon Masters enrich their D&D campaigns, or offerings to let parents customize Peppa Pig animations. “It’s supercharging fandom,” he says, “and I think that’s just net good for the brand.”
Cocks’ company attracted the attention of one AI platform owner last November, when a social media debate about whether D&D had gone “woke” prompted Elon Musk to ask, “How much is Hasbro?”
Cocks laughs off the tweet from the world’s richest man, saying any takeover interest went nowhere.
Met D&D in 1976 while in the Navy, learning Russian, and fell in love. We were steady for several years, then drifted apart and saw other people. Ran into each other a few years ago and the spark kinda rekindled, ya know? It’s an open relationship and I hang with GURPS and DragonQuest, but we’re all mutuals and it’s good. Open communication is the real key to any polydice, it’s the only real rule.
There might not be a formal definition, but when someone says "real AI" they typically are referring to a hypothetical actual intelligence like R2-D2 or Data, who are fully sapient beings capable of abstract reasoning and expressing their own desires and ideas. That's often what techbros sell AI as heading towards, too, even though it absolutely isn't.
Find your own truth, choose your enemies carefully, and never deal with a dragon.
"Canon" is what's factual to D&D lore. "Cannon" is what you're going to be shot with if you keep getting the word wrong.
Real AI is both artificial and, more to the point, intelligent. It doesn't have to be human level sapient or anything, just capable of understanding what it's communicating. A dog understands that when it barks, it is communicating to us that there is a potential intruder. A cat understands that if a mouse runs into a tube, it has to come out either end and will start planning accordingly.
What's called AI at the moment is not intelligent. It can trick you into thinking that it understands, but really it's just throwing words out that make sense. Sometimes. Thus it's not true AI.
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Intelligence is fundamentally about what's going on under the hood. If you can't accept discussions involving that, then the discussion is at a dead end.
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Ed Zitron has a great piece about how it's all a con to refer to things like ChatGPT as "AI."
And others about its impact on labour as well as on the environment.
The dissonance going on among those who claim to take protecting the latter seriously only to be cheering and applauding the development and use of generative AI is a dissonance becoming those in cults.
Nonsense. I didn't any such thing. You claimed the presence of intelligence is based on performance, it's not. That doesn't mean you have to dissect someone to know if they have intelligence.
I'm not sure whether you genuinely don't understand what's being said or if this has devolved into yet another "I must prove myself right at all costs!" thing, but the former will take a lot more effort that it's really worth to correct because it means going into a deep dive on what intelligence really is and the latter...is a boring conversation.
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Talking about what's going on 'under the hood' is fundamentally a dead end, because while we have some idea of what's going on on a fine detail level, at the holistic level required to evaluate whether something 'knows' what it's doing, we don't really know what's going on under the hood... for either humans or modern neural-net AIs. The reason we use observed behavior is because that's the information we have. I don't actually know that you're a thinking human being, I just assume that you are because you behave like one.
... and you'd only need to meet him / her in person and observe him / her to clarify what you suspect and—and this is important—what he / she knows.
Many are the works of philosophy and theology that explore what it means to be a 'thinking human being.' What you are saying about how this can only be 'observed' is not at all 'deep.' Because we are all conscious of this ourselves. What you are saying is analogous with saying some occurrence might not have occurred because no one was there to see it happen. Which is preposterous. It occurred or it didn't. An occurrence is not dependent on some observer's subjective experience of it. And someone's being a 'thinking human being' isn't dependent on it either. The universe doesn't revolve around us or around our awareness of it. We are aware of our belonging to the species to which we belong and of the fact it is a species with a capacity for intelligence. Science tells us that much. No amount of programming of mere language learning models is going to make them belong to our species. Or make them possess true intelligence. No matter how much we further ruin the environment in the process.